DEV Community

Bradley Harper
Bradley Harper

Posted on

I built an on-device AI face rater that runs 100% in the browser

Face-rating tools have always been a bad experience: upload a photo, wait for a server round-trip, sit behind a login wall, and get a watermarked single number. I wanted the opposite — a tool that measures facial geometry honestly, runs entirely on the device, and never asks for an account.

The result is FaceGrade, a free browser-based AI face rater. Here's how it works.

The model: 478 landmarks, on-device

Instead of calling a backend, the page loads MediaPipe's FaceLandmarker and runs it through WASM in the browser. One selfie produces 478 3D landmarks, and from those we derive the measurements people actually care about:

  • Facial symmetry — left vs. right landmark distances
  • Golden ratio — inter landmark proportions
  • Canthal tilt, jawline, midface and face shape
  • A 1–10 attractiveness score with a breakdown of what drives it

Nothing is uploaded. The image you pick is read locally and the landmarks are computed on your CPU/GPU, which is why the result appears in about a second and why there's no queue.

Why on-device matters here

Face photos are among the most sensitive images a person can upload. Every cloud-based rater has to ship that photo to a server, and most of them quietly train on it. Running the model locally means:

  • No account, no email, no watermark
  • No third-party tracking of your face
  • Works offline after the first load

The stack

It's a small TanStack Start app on Cloudflare Workers. The heavy part is purely client-side: MediaPipe tasks loaded from CDN, a custom score formula on top of the landmarks, and a downloadable report card so you can compare measurements over time.

If you've been using looksmaxxing-style tools that make you sign up, give it a try: FaceGrade — free AI face rater. Upload one photo and you'll get the full measurement breakdown in about a second.

Top comments (0)